Spaces:
Running
Running
| from io import BytesIO | |
| import numpy as np | |
| from PIL import Image | |
| IMAGE_SIZE = (32, 32) | |
| def extract_image_features(image_bytes): | |
| image = Image.open(BytesIO(image_bytes)).convert("RGB") | |
| image = image.resize(IMAGE_SIZE) | |
| pixels = np.asarray(image, dtype=np.float32) / 255.0 | |
| color_features = pixels.reshape(-1) | |
| histogram_features = np.concatenate([ | |
| np.histogram(pixels[:, :, channel], bins=16, range=(0, 1), density=True)[0] | |
| for channel in range(3) | |
| ]) | |
| grayscale = pixels.mean(axis=2) | |
| horizontal_edges = np.abs(np.diff(grayscale, axis=1)).mean() | |
| vertical_edges = np.abs(np.diff(grayscale, axis=0)).mean() | |
| texture_features = np.array([ | |
| grayscale.mean(), | |
| grayscale.std(), | |
| horizontal_edges, | |
| vertical_edges, | |
| ], dtype=np.float32) | |
| return np.concatenate([color_features, histogram_features, texture_features]) | |